A Markov model for predicting levels of psychiatric service use in borderline and antisocial personality disorders and bipolar type II affective disorder.

This study examines the relationship between borderline personality disorder (BPD) and the use of psychiatric services in a naturalistic follow-up comparison with antisocial personality disorder and bipolar type II affective disorder. In the first follow-up series, borderline psychopathology was associated with higher levels of psychiatric service use (emergency, daycare, and inpatient). Markov analyses indicated that the transition between levels of psychiatric service use followed a stationary, second order process (i.e. the immediate past and current service use, predicted use on the next follow-up, and the relationship did not depend on the point in time examined in the follow-up series). Further, the transition probabilities generated from this model did not depend significantly on diagnosis. Predictions from the Markov model about the cumulative probability that subjects would use the highest level of psychiatric services were tested on a second series of follow-ups on the same subjects 20 months later. The model-based predictions (starting from the observed levels in the first two follow-ups of the second series) clustered into three groups, of high, middle, and low predicted probabilities. The subject group with the lowest predicted likelihood had a cumulative probability of 0.19 for using emergency, daycare, or inpatient hospitalization by 22 months of follow-up, whereas the group with the highest likelihood (containing a disproportionate number of BPD subjects) had a cumulative observed probability of 0.80. The Markov model generated from this second series supported the stationarity of the transition process. BPD subjects began using high levels of psychiatric services, but their transition from one level to another over time followed a process similar to that of non-BPD subjects.